Non-Invasive Multistage Fruit Grading Application with User Recommendation system

In recent years, fruit sellers, consumers, and mid-lower income farmers have faced difficulty grading the fruits as it is laborious and needs massive investment. Artificial intelligence and vision sensors on mobile devices have led to non-invasive ways to grade the fruits. Hence, using deep learning, fruit grading applications with recommendation features were developed to handle multiple fruits. YoloV3 will detect the fruit type, followed by sub-categories classification using inceptionNet V3 and MobileNet V2 classifiers. Finally, Neural network classifier will predict the fruit grade based on handcrafted features. Deep neural network models were trained using two different data sets (i) fruit360 and (ii) our own (custom fruit dataset) in a transfer learning approach. The proposed application has client interface was developed using the angular framework, which communicates with the server using flask microservices. Where end-users can upload fruit images via mobile phones or web browsers to obtain (i) Fruit Sub Categories, and it grades with user recommendations such as (i) finding the nearest fruit shop (ii) Present retail market price of the fruit (iii) Recipe recommendation. The developed mobile application will remove bias and improve the perception of non-invasive fruit grading.